GHtest

Approximate test of equality of means when variances are heterogeneous.

Description

Games-Howell's approximate test of equality of means from normal population when variances are heterogeneous. It is a related test associated to the Behrens-Fisher problem.

It uses the Tukey's studentized range with specially weighted average degrees of freedom (df') and a standard error based on the averages of the variances of the means. Basically it realizes unplanned comparisions among pairs of means using the Games and Howell method. For the statistical test the function calls the file qTukey.m

File needs to input the data matrix [1=yes (default); 2=no (If not, you must to give the statistics matrix] and the significance level (default = 0.05).

It Outputs a complete table of the statistical analysis of the difference between each pair of means.

• MATLAB release: MATLAB 5.3 (R11)

[Michael Volles]

I compared the output of this program with the GH test in SPSS, and found significant differences. Looking at the m file, I believe the studentized range (qtukey)needs to be divided by sqrt(2)). This corrected the discrepancy with SPSS and is consistent with Hochberg and Tamhane 1987 mult. comparison procedures. Correct me if I'm wrong.

[Alessandro Corrêa]

It's a very useful test, once homogeneous variances are seldom found in economomics and financial markets.
The M-file is well built and help text is clear